(583fy) Simplex and Model Based Self-Optimization With In-Situ-FTIR Analytics

In the last few years, self-optimizing reactors with numeric algorithms are a growing field in organic chemistry publications. With different methods like Simplex, Snobfit or Steepest Descent Method, different reactions were optimized with respect to yield, space time yield, E-factor and other criteria. To use these algorithms in continuous reactions systems, it is necessary to use an online analytic like HPLC, GLC or FTIR-spectroscopy.

For the use of FTIR-spectroscopy in systems with complex IR-spectra, a quantification model is needed. Here it is important to use an easy and fast method to build up this model. With a time-consuming quantification method, the advantage of an optimization with Simplex is lost.

The Simplex method comes in three different main variations (Simplex, Modified Simplex and Super Modified Simplex) and many further modifications. During an optimization run with these algorithms, different measuring points are measured. However, in N-dimensional parameter space, the Simplex algorithm only uses (N + 1) points in every optimization step. Prior points are discarded. This non-use of information is a disadvantage of the Simplex algorithms.

Aims

The aim of this work is the use of a Modified Simplex algorithm and a simple, kinetic model. The integration of a kinetic model should improve the speed of optimization. In addition, it adds the function to find the global optimum instead of a local one with only the Simplex method.

Methods

For the proof of the presented optimization method, a simple reaction setup is used (see figure). The substrates are pumped into a coiled tubular reactor with two HPLC pumps. The reactor temperature is controlled with an oil bath. With this setup it is possible to optimize the reaction in three dimensions; residence time, concentration and temperature. The optimization algorithm is controlled by computer with LabView and Matlab.

The online analytic is done via two in-situ-FTIR flow cells. One cell is placed before the reactor and one at the outlet. With this kind of spectroscopy it is possible to follow the reaction progress with high time resolution. Because of that the steady state of the reaction is detected faster than with the chromatographic method.

To compare the difference of the Simplex algorithm with the model based method, a reaction is optimized twice, with every method once. The aim for a reaction to prove the presented concept is a more complex and more industrially relevant system like in literature to combine the integration of an IR-quantification model and the new optimization method.

Results

The implementation of a Modified Simplex algorithm with FTIR analytics was done and tested with a two dimensional system which was optimized to 97% conversion through residence time and substrate ratio.

For the implementation of the kinetic model a theoretical investigation of a model reaction was carried out. A consecutive reaction of component A with component B to product C, and of C with B to side product D was used. The optimization with this method showed a significant increase of optimization speed with respect to a straight Simplex, because it reduced the needed steps to the optimum.

With these promising results the experimental proof will exhibit the potential of the presented optimization method.